Meta
Llama 4 Maverick
Open-weight model with 400 billion parameters, released with the smaller Llama 4 Scout (109 billion). Llama 4 Behemoth was previewed but never released.
- Announced
- Access
- Public
Benchmark scores
| Test | Score | Best on the tracker | How it was measured |
|---|---|---|---|
| Humanity's Last Examexpert questions, no tools | 5.7% | GPT-6 Astra 55% | Independentleaderboard entry "Llama 4 Maverick" (Llama-4-Maverick-17B-128E-Instruct) |
| GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated | 67% | Gemini 4 Argon 96% | Independentdefault settings (Llama-4-Maverick-17B-128E-Instruct-FP8)Run 8 Apr 2025 |
| ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated | 0% | GPT-6 Astra 95% | IndependentARC Prize config "Llama 4 Maverick" (Llama-4-Maverick-17B-128E-Instruct-FP8-together) |
| OTIS Mock AIMEcompetition mathsSaturated | 21% | GPT-5.5 100% | Independentdefault settings (Llama-4-Maverick-17B-128E-Instruct-FP8)Run 8 Apr 2025 |
Sources
- The Llama 4 herdMeta AI · Official · ai.meta.com
- meta-llama on Hugging FaceHugging Face · Dataset · huggingface.co
- Epoch AI Benchmarking Hub (model release dates and runs)Epoch AI · Dataset · 9 Nov 2026 · epoch.ai